hedwigaihedwigaihedwigai

Experiment with perspectives

Choose one of the 13 perspective presets. Or push a neuron of your own to make a new perspective. Then ask a question. You will see the model’s plain answer next to the new one. The post explains how it works.

1. Pick a neuron

2. Mod the neuron

push =1.00× median‖h‖ ÷ ‖Wdown[:,]‖
Strength

Added to the neuron at every token. The two norms scale it so that 1 adds as much as the layer already carries. Lower keeps more of the plain answer; higher breaks the model sooner.

3. Ask it something

Sign in to run

New accounts get free runs; after that, runs are billed to your credit.

4. Compare

Plain

The model, nothing pushed

Pick a neuron, ask it something, and press the button.

Pushed

The same model, one neuron pushed

Pick a neuron, ask it something, and press the button.